Research On The Medical Image Fusion Algorithm Based On Non-subsampled Shearlet Transform

Image fusion technology can integrate the complementary information of medical images of various modes into the same image,and the integrated image information is comprehensive and rich in content.It can describe the lesion location more accurately and in detail,and improve the accuracy of doctors’ diagnosis and treatment.This thesis mainly studies the medical image fusion algorithm based on non-subsampled shearlet transform(NSST),the research content and innovation are as follows:1.In order to improve the fusion accuracy of CT and MRI images and retain more anatomical information of source images,a medical image fusion algorithm based on NSST and parameter-adaptive pulse coupled neural network(PA-PCNN)model was proposed.In the proposed algorithm,the NSST de-composition is first performed on the source images in multi-scale and multi-direction to obtain one low-frequency sub-band and a series of high-frequency sub-bands.Secondly,a PA-PCNN model is used to fuse the high frequency sub-bands.PCNN model has been proved to be very suitable for image fusion and can preserve the texture and detail information of the source images.However,the traditional PCNN model usually needs to set a series of free parameters according to the experimental experience or results,so that the relationship between the final results and parameters of image processing is not clear.PA-PCNN model not only solves the problem of manual setting of free parameters,increases the correlation between pixels,but also better retain the texture information of the image.For the fusion of low frequency sub-bands,the traditional low frequency fusion rules based on average often lead to energy loss in fused images.Therefore,an energy attribute(EA)fusion rule is more conducive to the preservation of complete basic information.Finally,the fused image is reconstructed by performing inverse NSST.The experimental results demonstrate that the fused images obtained in this thesis have clear contours,high contrast,better preservation of detailed texture,effectively avoid the emergence of artifacts and blur,and better results in objective indicators.2.In order to improve the fusion accuracy of MRI,PET and SPECT images,the fused medical images contain more functional and anatomical information,better distinguish between diseased tissues and normal tissues.A medical image fusion algorithm based on NSST and simplified PCNN(SPCNN)model is proposed.Firstly,the source image is transformed by NSST in multi-scale and multi-direction,and the high frequency sub-bands are fused by a SPCNN model.In a series of free parameters of SPCNN model,link strength is a very key parameter,which has a great influence on the final image fusion quality.Adaptive adjustment of link strength can effectively improve the performance of image fusion.Therefore,multi-scale morphological gradient is introduced.Link strength in SPCNN model is adjusted by multi-scale morphological gradient to enhance image spatial correlation.For the fusion of low frequency sub-bands,the traditional low-frequency fusion rules based on average are easy to lose the energy information of the source images,resulting in the brightness of some areas decreases rapidly and the color fidelity is not high.Therefore,the fusion of low frequency sub-bands adopt a fusion rule based on the combination of weighted local energy(WLE)and weighted sum of eight-neighborhood-based modified laplacian(WSEML).It better preserves the image energy and details.Finally,the fused image is reconstructed by performing inverse NSST.The experimental results demonstrate that the fused images obtained in this thesis has a good visual effect,effectively retains the edge texture details of the source images and has a better color fidelity,and a better effect is obtained in the objective indicators.